A naukri.com initiative
ML News
Arxiv
3d
204
Image Credit: Arxiv
Investigating Relational State Abstraction in Collaborative MARL
- This paper investigates the impact of relational state abstraction on sample efficiency and performance in collaborative Multi-Agent Reinforcement Learning (MARL).
- The proposed abstraction is based on spatial relationships, leveraging spatial reasoning in real-world multi-agent scenarios.
- The authors introduce MARC (Multi-Agent Relational Critic), a critic architecture that incorporates spatial relational inductive biases by transforming the state into a spatial graph and processing it through a relational graph neural network.
- Empirical analysis shows that MARC outperforms state-of-the-art MARL baselines in terms of sample efficiency, asymptotic performance, and potential for generalization, without requiring complex designs or task-specific engineering.
Read Full Article
12 Likes
Arxiv
3d
164
Image Credit: Arxiv
Cosmology with Persistent Homology: Parameter Inference via Machine Learning
- This article investigates the potential of persistent homology for constraining cosmological parameters and primordial non-Gaussianity amplitudes.
- Persistent homology using persistence images (PIs) performs better than the combined Power Spectrum and Bispectrum (PS/BS) for inferring parameters.
- PIs show promise in constraining primordial non-Gaussianity, particularly for the parameter fNL^loc.
- The combination of PIs with PS/BS provides only marginal gains, indicating little extra information in PS/BS compared to PIs.
Read Full Article
9 Likes
Arxiv
3d
84
Image Credit: Arxiv
Energy consumption of code small language models serving with runtime engines and execution providers
- The study focused on analyzing the impact of deep learning runtime engines and execution providers on energy consumption, execution time, and computing-resource utilization in the context of code Small Language Models (SLMs).
- CUDA execution provider configurations outperformed CPU execution provider configurations in terms of energy consumption and execution time.
- TORCH paired with CUDA demonstrated the greatest energy efficiency, achieving energy savings from 37.99% up to 89.16% compared to other serving configurations.
- Optimized runtime engines like ONNX with the CPU execution provider achieved from 8.98% up to 72.04% energy savings within CPU-based configurations.
Read Full Article
5 Likes
Arxiv
3d
345
Image Credit: Arxiv
Learning charges and long-range interactions from energies and forces
- Accurate modeling of long-range forces is crucial in atomistic simulations for understanding material properties.
- Standard machine learning interatomic potentials often rely on short-range approximations, limiting their applicability in systems with significant electrostatics and dispersion forces.
- The Latent Ewald Summation (LES) method was introduced to capture long-range electrostatics without explicitly learning atomic charges or charge equilibration.
- LES has been successfully applied in benchmarking various challenging systems, showing its effectiveness in inferring physical partial charges, dipole and quadrupole moments, and achieving higher accuracy compared to methods that explicitly learn charges.
Read Full Article
20 Likes
Arxiv
3d
232
Image Credit: Arxiv
TalkWithMachines: Enhancing Human-Robot Interaction for Interpretable Industrial Robotics Through Large/Vision Language Models
- TalkWithMachines aims to enhance human-robot interaction in interpretable industrial robotics.
- The paper explores the integration of Large Language Models (LLMs) and Vision Language Models (VLMs) with robotic perception and control.
- This enables robots to understand and execute commands in natural language and perceive their environment through visual and descriptive inputs.
- The research focuses on four LLM-assisted simulated robotic control workflows, including low-level control, language-based feedback, visual information usage, and task planning.
Read Full Article
14 Likes
Arxiv
3d
104
Image Credit: Arxiv
Predicting Long-Term Student Outcomes from Short-Term EdTech Log Data
- Educational stakeholders are interested in sparse, delayed student outcomes like end-of-year statewide exams.
- Prior work has focused on using long-term usage data to predict outcomes, but this study investigates using short-term log data to predict students' end-of-school year assessments.
- The study utilizes datasets from students in Uganda using a literacy game product and students in the US using two mathematics tutoring systems.
- Findings suggest that 2-5 hours of log usage data can provide valuable insight into students' long-term performance.
Read Full Article
6 Likes
Arxiv
3d
261
Image Credit: Arxiv
DualGFL: Federated Learning with a Dual-Level Coalition-Auction Game
- Researchers propose DualGFL, a novel Federated Learning framework with a Dual-level Game in cooperative-competitive environments.
- DualGFL includes a lower-level hedonic game where clients form coalitions and an upper-level multi-attribute auction game where coalitions bid for training participation.
- At the lower-level, DualGFL introduces a new auction-aware utility function and a Pareto-optimal partitioning algorithm to find a Pareto-optimal partition based on clients' preference profiles.
- At the upper-level, DualGFL formulates a multi-attribute auction game with resource constraints and derives equilibrium bids to maximize coalitions' winning probabilities and profits.
Read Full Article
15 Likes
Arxiv
3d
345
Image Credit: Arxiv
The Impact of Cut Layer Selection in Split Federated Learning
- Split Federated Learning (SFL) combines federated learning and split learning.
- SFL partitions a neural network at a cut layer, with initial layers on clients and remaining layers on a training server.
- SFL-V1 maintains separate server-side models for each client, while SFL-V2 maintains a single shared model for all clients.
- Cut layer selection significantly affects the performance of SFL-V2, outperforming FedAvg on certain datasets.
Read Full Article
20 Likes
Arxiv
3d
216
Image Credit: Arxiv
NGQA: A Nutritional Graph Question Answering Benchmark for Personalized Health-aware Nutritional Reasoning
- Researchers have introduced the Nutritional Graph Question Answering (NGQA) benchmark
- NGQA is the first graph question answering dataset designed for personalized nutritional health reasoning
- The benchmark leverages data from the National Health and Nutrition Examination Survey (NHANES) and the Food and Nutrient Database for Dietary Studies (FNDDS)
- NGQA effectively challenges existing models and advances GraphQA research with a novel domain-specific benchmark
Read Full Article
13 Likes
Arxiv
3d
397
Image Credit: Arxiv
Predicting Artificial Neural Network Representations to Learn Recognition Model for Music Identification from Brain Recordings
- Recent studies have shown that artificial neural network (ANN) representations can resemble cortical representations when exposed to the same auditory inputs.
- This study proposes a new approach by using ANN representations as a supervisory signal to train recognition models for music identification using non-invasive brain recordings.
- By training an EEG recognition model to predict ANN representations associated with music identification, significant improvement in classification accuracy is observed.
- This research has potential implications for advancing brain-computer interfaces, neural decoding techniques, and our understanding of music cognition.
Read Full Article
23 Likes
Arxiv
3d
385
Image Credit: Arxiv
In-context Continual Learning Assisted by an External Continual Learner
- Existing continual learning (CL) methods rely on fine-tuning or adapting large language models (LLMs) but suffer from catastrophic forgetting (CF).
- In-context learning (ICL) can leverage the extensive knowledge within LLMs for CL without updating any parameters.
- However, scaling ICL becomes challenging as the prompt length increases and exceeds the input token limit.
- To address this, the InCA approach integrates an external continual learner (ECL) with ICL, resulting in scalable CL without CF and achieving significant performance gains.
Read Full Article
23 Likes
Arxiv
3d
261
Image Credit: Arxiv
Multi Agent Reinforcement Learning for Sequential Satellite Assignment Problems
- Assignment problems are combinatorial optimization problems where agents need to be assigned to tasks while maximizing utility and satisfying constraints.
- Multi-agent reinforcement learning (MARL) is applied to solve assignment problems that unfold over time.
- The algorithm uses bootstrapping from a polynomial-time greedy solver and further experience to learn the value of assignments.
- The distributed optimal assignment mechanism is employed to choose assignments.
Read Full Article
15 Likes
Arxiv
3d
245
Image Credit: Arxiv
SaliencyI2PLoc: saliency-guided image-point cloud localization using contrastive learning
- SaliencyI2PLoc is a contrastive learning based architecture for image to point cloud global localization.
- It addresses the modality gap between images and point clouds and maintains feature alignment and relation consistency.
- Saliency map is fused into feature aggregation for more representative global features.
- Experiments show significant improvement in cross-modality global localization compared to baseline methods.
Read Full Article
14 Likes
Arxiv
3d
236
Image Credit: Arxiv
Score-based Generative Diffusion Models for Social Recommendations
- Score-based Generative Diffusion Models for Social Recommendations
- This paper addresses the challenge of low social homophily in social recommendations.
- They propose the Score-based Generative Model for Social Recommendation (SGSR) which adapts Stochastic Differential Equation (SDE) based diffusion models.
- Experiments show that SGSR effectively filters redundant social information and improves recommendation performance.
Read Full Article
14 Likes
Arxiv
3d
120
Image Credit: Arxiv
Dexterous Manipulation Based on Prior Dexterous Grasp Pose Knowledge
- Dexterous manipulation has been a focus of recent research.
- Existing studies have primarily used reinforcement learning methods for hand movements, but these methods are often inefficient and inaccurate.
- This work introduces a novel reinforcement learning approach that utilizes prior dexterous grasp pose knowledge to improve efficiency and accuracy.
- The manipulation process is divided into two phases: generating a dexterous grasp pose targeting the functional part of the object, and using reinforcement learning to explore the environment.
Read Full Article
7 Likes
For uninterrupted reading, download the app